LiveBaster - self-sufficient Artificial Intelligence (AI)

на Русский

Product

   Video

   Applications

   Buy

Documentation

Contacts

Purpose

   Self-learning Artificial Intelligence (SAI) is designed for multichannel management of various objects in a fully autonomous mode. An object is a device (or computer animation), which has several control channels and multiple sensor signals. The object may be in the form of a single object (a robot) and in the form of a set of objects (a group of robots). With some modification, the SAI can also be used to recognize images or sounds. SAI is universal, it can control a robot of any type, with any number of limbs, it can be legs, wheels, etc. In case of breakage or partial breakage of the device, the SAI self-teaches to adjust in motion (similar to a living creature that hurt a foot).

Main characteristic

   A device is a software base component (hereinafter referred to as a base component or a BC), originally written in C++, a compiled base component is supplied with a DLL extension for Windows. BC is unified, mainly uses the basic functions of the operating system, so it does not depend on the version of the operating system, there are modules for 32 and 64-bit OS. The basic component operates in real time, can manage various objects with different numbers of control channels (up to 100) depending on the performance of the processor and the memory resource. The standard desktop computer with I5 processor and 4GB memory is capable of providing about 20 control channels in real time. It is possible to use the SAI in smartphones and tablets with these operating systems, the release for IOS is also planned. The size of the BC (software module kit) is not more than 50Mb (excluding the size of the knowledge base, the size of the knowledge base depends on the task, the number of control channels, the accumulated experience of the module). The maximum knowledge base size is 1 GB.

Principles of operation

   The basic component does not require pre-training, training starts at the same time at the start of the program, and training is conducted continuously throughout the life of the SAI. It is an imitation of the work of a number of divisions of the vertebrate brain and is trained on the same principles as a real animal. The currently existing on the market, developed by various manufacturers artificial intelligence (AI) is based on logical processing of information by various algorithms (based on the use of a decision algorithm, or using neural networks, etc.). To describe the properties of AI, it is possible to draw an analogy with higher nervous activity, or the work of the second signal system (these are not quite equivalent terms, but within this approach, the difference can be ignored). Unlike conventional AI, our module is called SAI, as it simulates the operation of the primary signal system based on reflexes and subject to continuous self-learning. For tasks that an animal performs in nature (identification and movement), the SAI is optimal.

Purpose of use

   To use the software package, you must first connect the base component (BC) to all prospective actuators and sensors. BC can be used as part of a program (for example, C++ or C#), and the structure of the transmitted data and the method of handling are given in the next section. This method may be required when controlling a real robot (industrial or toy). Thus it is necessary to consider that at the beginning of work the SAI has absolutely no idea of properties of the managed object, but absolutely any commands can be submitted. This is similar to the movement of a newborn animal (e.g., foal). In the case of a real robot, this could lead to breakage or destruction of the robot drives, so that this does not happen it is necessary to provide mechanical protection for important units and sensors so that the robot does not cause itself harm. The protection can be software-based, in which case the commands for the actuators must be checked for compatibility with a specific robot device. In practice, this is quite difficult to do, since it is impossible to predict all combinations of control commands for a complex robot, so mechanical protection is recommended even if there is software one. It should also be noted that even the preliminary training of the robot can not serve as a guarantee of the absence of harm to the robot itself, as it is continuously trained and its actions are as unpredictable as the actions of an ordinary animal.

© 1993-2021 Victor Artyukhov